One audio model now spans speech, music, and sound effects
Most audio generators still specialize. StepAudio 3 Gen instead uses one discrete autoregressive model to produce zero-shot speech, designed voices, vocals, music, sound effects, and mixtures of them. Its shared RVQ representation and 16-codebook pipeline reportedly reach state-of-the-art results in text-to-speech and voice design, while making mixed-audio generation a first-class capability rather than a stitching problem.